attention head AI jobs Alerts Saved
AI jobs › Safety & interpretability AI jobs

Safety & interpretability AI jobs

Safety work tries to make sure models do what people intend and can't easily be turned to harm. It covers alignment research on training models whose goals and behaviour stay as intended as they grow more capable, evaluations of dangerous capabilities, safeguards and classifiers that catch misuse, and red-teaming: attacking a model on purpose, with jailbreaks and other tricks, to find its failures before others do.

Interpretability studies what happens inside a network: what its activations represent and how they combine into the model's behaviour, so that claims about a model can rest on more than its outputs. Many of these roles are research scientists and research engineers at labs and research organisations, with policy, trust and safety, and security roles around them. Expect strong ML fundamentals, PyTorch, careful experiments, and clear writing about what results do and don't show.

We count roles whose postings mention AI safety, model safety, safety research, alignment research, science or team, interpretability, red-teaming, or jailbreaks. Red team is also a security term for attacking an organisation's own systems, and interpretability can mean explaining a simpler model's predictions, so some security and data science roles are counted here too.

283 open AI roles at 142 employers, from their own job boards. Showing the newest 100.

Follow Safety & interpretability37 new this week

New Safety & interpretability roles by email, the morning after we find them, in one email with your other alerts. Narrow it, or keep every role.

How often
We email a link to confirm it.
OpenAI · 20FAR.AI · 19Faculty · 14Anthropic · 13Lila Sciences · 1110a Labs · 6Microsoft · 6Cambridge Boston Alignment Initiative · 5NVIDIA · 5Bosch Group · 4Carnegie Mellon University · 4ServiceNow · 4Singapore AI Safety Hub · 4Amazon · 3Capco · 3

Get new Safety & interpretability roles every Monday

A job alert: new Safety & interpretability roles from employers' own job boards, every Monday. Unsubscribe in one click.

Related